Oracle Failure
When the outside data a blockchain application depends on, usually a price, is wrong, delayed, or manipulated, and the application acts on it anyway.
Blockchains cannot see outside themselves, so lending protocols, derivatives, and stablecoins import prices through oracles. If an oracle reads a single venue or a thin liquidity pool, an attacker can move that price temporarily, often with borrowed funds repaid in the same transaction through a flash loan, and then borrow against collateral the protocol now overvalues or trigger liquidations that should not occur. Failures also come from staleness, where a feed stops updating during volatility, and from mismatched decimals or an incorrect asset mapping in configuration. Mitigations include aggregating many independent sources, time-weighted averages that are expensive to move, deviation thresholds, heartbeat checks, and circuit breakers that pause a market when the feed looks implausible.
In practice
A lending market that prices collateral from one automated market maker pool can be attacked by pushing that pool's price with a large temporary trade, borrowing against the inflated valuation, and leaving the debt behind.
The common misunderstanding
That an oracle failure means the blockchain was hacked, when the chain executed exactly as written and the fault was in the data it was given.